Code Llama Instruct 34B accepts only 4096 tokens on PRO

The PRO page Inference for PROs mentions that the model Code Llama Instruct 34B has a context length of 16K. On trying to inference with a PRO account using 4900 tokens, it threw an error mentioning limit of tokens is 4096.

Is this expected behavior? Shouldn’t this be clearly mentioned in the PRO page before we subscribe and pay for it.